2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/seq_file.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/moduleparam.h>
18 #include <linux/parser.h>
19 #include <linux/statfs.h>
20 #include <linux/random.h>
21 #include <linux/sched.h>
22 #include <linux/exportfs.h>
24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
25 MODULE_DESCRIPTION("Filesystem in Userspace");
26 MODULE_LICENSE("GPL");
28 static struct kmem_cache
*fuse_inode_cachep
;
29 struct list_head fuse_conn_list
;
30 DEFINE_MUTEX(fuse_mutex
);
32 static int set_global_limit(const char *val
, struct kernel_param
*kp
);
34 unsigned max_user_bgreq
;
35 module_param_call(max_user_bgreq
, set_global_limit
, param_get_uint
,
36 &max_user_bgreq
, 0644);
37 __MODULE_PARM_TYPE(max_user_bgreq
, "uint");
38 MODULE_PARM_DESC(max_user_bgreq
,
39 "Global limit for the maximum number of backgrounded requests an "
40 "unprivileged user can set");
42 unsigned max_user_congthresh
;
43 module_param_call(max_user_congthresh
, set_global_limit
, param_get_uint
,
44 &max_user_congthresh
, 0644);
45 __MODULE_PARM_TYPE(max_user_congthresh
, "uint");
46 MODULE_PARM_DESC(max_user_congthresh
,
47 "Global limit for the maximum congestion threshold an "
48 "unprivileged user can set");
50 #define FUSE_SUPER_MAGIC 0x65735546
52 #define FUSE_DEFAULT_BLKSIZE 512
54 /** Maximum number of outstanding background requests */
55 #define FUSE_DEFAULT_MAX_BACKGROUND 12
57 /** Congestion starts at 75% of maximum */
58 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
60 struct fuse_mount_data
{
65 unsigned fd_present
:1;
66 unsigned rootmode_present
:1;
67 unsigned user_id_present
:1;
68 unsigned group_id_present
:1;
74 struct fuse_forget_link
*fuse_alloc_forget(void)
76 return kzalloc(sizeof(struct fuse_forget_link
), GFP_KERNEL
);
79 static struct inode
*fuse_alloc_inode(struct super_block
*sb
)
82 struct fuse_inode
*fi
;
84 inode
= kmem_cache_alloc(fuse_inode_cachep
, GFP_KERNEL
);
88 fi
= get_fuse_inode(inode
);
96 INIT_LIST_HEAD(&fi
->write_files
);
97 INIT_LIST_HEAD(&fi
->queued_writes
);
98 INIT_LIST_HEAD(&fi
->writepages
);
99 init_waitqueue_head(&fi
->page_waitq
);
100 fi
->forget
= fuse_alloc_forget();
102 kmem_cache_free(fuse_inode_cachep
, inode
);
109 static void fuse_i_callback(struct rcu_head
*head
)
111 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
112 kmem_cache_free(fuse_inode_cachep
, inode
);
115 static void fuse_destroy_inode(struct inode
*inode
)
117 struct fuse_inode
*fi
= get_fuse_inode(inode
);
118 BUG_ON(!list_empty(&fi
->write_files
));
119 BUG_ON(!list_empty(&fi
->queued_writes
));
121 call_rcu(&inode
->i_rcu
, fuse_i_callback
);
124 static void fuse_evict_inode(struct inode
*inode
)
126 truncate_inode_pages(&inode
->i_data
, 0);
128 if (inode
->i_sb
->s_flags
& MS_ACTIVE
) {
129 struct fuse_conn
*fc
= get_fuse_conn(inode
);
130 struct fuse_inode
*fi
= get_fuse_inode(inode
);
131 fuse_queue_forget(fc
, fi
->forget
, fi
->nodeid
, fi
->nlookup
);
136 static int fuse_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
138 if (*flags
& MS_MANDLOCK
)
145 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
146 * so that it will fit.
148 static ino_t
fuse_squash_ino(u64 ino64
)
150 ino_t ino
= (ino_t
) ino64
;
151 if (sizeof(ino_t
) < sizeof(u64
))
152 ino
^= ino64
>> (sizeof(u64
) - sizeof(ino_t
)) * 8;
156 void fuse_change_attributes_common(struct inode
*inode
, struct fuse_attr
*attr
,
159 struct fuse_conn
*fc
= get_fuse_conn(inode
);
160 struct fuse_inode
*fi
= get_fuse_inode(inode
);
162 fi
->attr_version
= ++fc
->attr_version
;
163 fi
->i_time
= attr_valid
;
165 inode
->i_ino
= fuse_squash_ino(attr
->ino
);
166 inode
->i_mode
= (inode
->i_mode
& S_IFMT
) | (attr
->mode
& 07777);
167 set_nlink(inode
, attr
->nlink
);
168 inode
->i_uid
= make_kuid(&init_user_ns
, attr
->uid
);
169 inode
->i_gid
= make_kgid(&init_user_ns
, attr
->gid
);
170 inode
->i_blocks
= attr
->blocks
;
171 inode
->i_atime
.tv_sec
= attr
->atime
;
172 inode
->i_atime
.tv_nsec
= attr
->atimensec
;
173 inode
->i_mtime
.tv_sec
= attr
->mtime
;
174 inode
->i_mtime
.tv_nsec
= attr
->mtimensec
;
175 inode
->i_ctime
.tv_sec
= attr
->ctime
;
176 inode
->i_ctime
.tv_nsec
= attr
->ctimensec
;
178 if (attr
->blksize
!= 0)
179 inode
->i_blkbits
= ilog2(attr
->blksize
);
181 inode
->i_blkbits
= inode
->i_sb
->s_blocksize_bits
;
184 * Don't set the sticky bit in i_mode, unless we want the VFS
185 * to check permissions. This prevents failures due to the
186 * check in may_delete().
188 fi
->orig_i_mode
= inode
->i_mode
;
189 if (!(fc
->flags
& FUSE_DEFAULT_PERMISSIONS
))
190 inode
->i_mode
&= ~S_ISVTX
;
192 fi
->orig_ino
= attr
->ino
;
195 void fuse_change_attributes(struct inode
*inode
, struct fuse_attr
*attr
,
196 u64 attr_valid
, u64 attr_version
)
198 struct fuse_conn
*fc
= get_fuse_conn(inode
);
199 struct fuse_inode
*fi
= get_fuse_inode(inode
);
201 struct timespec old_mtime
;
203 spin_lock(&fc
->lock
);
204 if (attr_version
!= 0 && fi
->attr_version
> attr_version
) {
205 spin_unlock(&fc
->lock
);
209 old_mtime
= inode
->i_mtime
;
210 fuse_change_attributes_common(inode
, attr
, attr_valid
);
212 oldsize
= inode
->i_size
;
213 i_size_write(inode
, attr
->size
);
214 spin_unlock(&fc
->lock
);
216 if (S_ISREG(inode
->i_mode
)) {
219 if (oldsize
!= attr
->size
) {
220 truncate_pagecache(inode
, oldsize
, attr
->size
);
222 } else if (fc
->auto_inval_data
) {
223 struct timespec new_mtime
= {
224 .tv_sec
= attr
->mtime
,
225 .tv_nsec
= attr
->mtimensec
,
229 * Auto inval mode also checks and invalidates if mtime
232 if (!timespec_equal(&old_mtime
, &new_mtime
))
237 invalidate_inode_pages2(inode
->i_mapping
);
241 static void fuse_init_inode(struct inode
*inode
, struct fuse_attr
*attr
)
243 inode
->i_mode
= attr
->mode
& S_IFMT
;
244 inode
->i_size
= attr
->size
;
245 if (S_ISREG(inode
->i_mode
)) {
246 fuse_init_common(inode
);
247 fuse_init_file_inode(inode
);
248 } else if (S_ISDIR(inode
->i_mode
))
249 fuse_init_dir(inode
);
250 else if (S_ISLNK(inode
->i_mode
))
251 fuse_init_symlink(inode
);
252 else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
253 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
254 fuse_init_common(inode
);
255 init_special_inode(inode
, inode
->i_mode
,
256 new_decode_dev(attr
->rdev
));
261 int fuse_inode_eq(struct inode
*inode
, void *_nodeidp
)
263 u64 nodeid
= *(u64
*) _nodeidp
;
264 if (get_node_id(inode
) == nodeid
)
270 static int fuse_inode_set(struct inode
*inode
, void *_nodeidp
)
272 u64 nodeid
= *(u64
*) _nodeidp
;
273 get_fuse_inode(inode
)->nodeid
= nodeid
;
277 struct inode
*fuse_iget(struct super_block
*sb
, u64 nodeid
,
278 int generation
, struct fuse_attr
*attr
,
279 u64 attr_valid
, u64 attr_version
)
282 struct fuse_inode
*fi
;
283 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
286 inode
= iget5_locked(sb
, nodeid
, fuse_inode_eq
, fuse_inode_set
, &nodeid
);
290 if ((inode
->i_state
& I_NEW
)) {
291 inode
->i_flags
|= S_NOATIME
|S_NOCMTIME
;
292 inode
->i_generation
= generation
;
293 inode
->i_data
.backing_dev_info
= &fc
->bdi
;
294 fuse_init_inode(inode
, attr
);
295 unlock_new_inode(inode
);
296 } else if ((inode
->i_mode
^ attr
->mode
) & S_IFMT
) {
297 /* Inode has changed type, any I/O on the old should fail */
298 make_bad_inode(inode
);
303 fi
= get_fuse_inode(inode
);
304 spin_lock(&fc
->lock
);
306 spin_unlock(&fc
->lock
);
307 fuse_change_attributes(inode
, attr
, attr_valid
, attr_version
);
312 int fuse_reverse_inval_inode(struct super_block
*sb
, u64 nodeid
,
313 loff_t offset
, loff_t len
)
319 inode
= ilookup5(sb
, nodeid
, fuse_inode_eq
, &nodeid
);
323 fuse_invalidate_attr(inode
);
325 pg_start
= offset
>> PAGE_CACHE_SHIFT
;
329 pg_end
= (offset
+ len
- 1) >> PAGE_CACHE_SHIFT
;
330 invalidate_inode_pages2_range(inode
->i_mapping
,
337 static void fuse_umount_begin(struct super_block
*sb
)
339 fuse_abort_conn(get_fuse_conn_super(sb
));
342 static void fuse_send_destroy(struct fuse_conn
*fc
)
344 struct fuse_req
*req
= fc
->destroy_req
;
345 if (req
&& fc
->conn_init
) {
346 fc
->destroy_req
= NULL
;
347 req
->in
.h
.opcode
= FUSE_DESTROY
;
350 fuse_request_send(fc
, req
);
351 fuse_put_request(fc
, req
);
355 static void fuse_bdi_destroy(struct fuse_conn
*fc
)
357 if (fc
->bdi_initialized
)
358 bdi_destroy(&fc
->bdi
);
361 void fuse_conn_kill(struct fuse_conn
*fc
)
363 spin_lock(&fc
->lock
);
367 spin_unlock(&fc
->lock
);
368 /* Flush all readers on this fs */
369 kill_fasync(&fc
->fasync
, SIGIO
, POLL_IN
);
370 wake_up_all(&fc
->waitq
);
371 wake_up_all(&fc
->blocked_waitq
);
372 wake_up_all(&fc
->reserved_req_waitq
);
374 EXPORT_SYMBOL_GPL(fuse_conn_kill
);
376 static void fuse_put_super(struct super_block
*sb
)
378 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
380 fuse_send_destroy(fc
);
383 mutex_lock(&fuse_mutex
);
384 list_del(&fc
->entry
);
385 fuse_ctl_remove_conn(fc
);
386 mutex_unlock(&fuse_mutex
);
387 fuse_bdi_destroy(fc
);
392 static void convert_fuse_statfs(struct kstatfs
*stbuf
, struct fuse_kstatfs
*attr
)
394 stbuf
->f_type
= FUSE_SUPER_MAGIC
;
395 stbuf
->f_bsize
= attr
->bsize
;
396 stbuf
->f_frsize
= attr
->frsize
;
397 stbuf
->f_blocks
= attr
->blocks
;
398 stbuf
->f_bfree
= attr
->bfree
;
399 stbuf
->f_bavail
= attr
->bavail
;
400 stbuf
->f_files
= attr
->files
;
401 stbuf
->f_ffree
= attr
->ffree
;
402 stbuf
->f_namelen
= attr
->namelen
;
403 /* fsid is left zero */
406 static int fuse_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
408 struct super_block
*sb
= dentry
->d_sb
;
409 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
410 struct fuse_req
*req
;
411 struct fuse_statfs_out outarg
;
414 if (!fuse_allow_current_process(fc
)) {
415 buf
->f_type
= FUSE_SUPER_MAGIC
;
419 req
= fuse_get_req_nopages(fc
);
423 memset(&outarg
, 0, sizeof(outarg
));
425 req
->in
.h
.opcode
= FUSE_STATFS
;
426 req
->in
.h
.nodeid
= get_node_id(dentry
->d_inode
);
427 req
->out
.numargs
= 1;
428 req
->out
.args
[0].size
=
429 fc
->minor
< 4 ? FUSE_COMPAT_STATFS_SIZE
: sizeof(outarg
);
430 req
->out
.args
[0].value
= &outarg
;
431 fuse_request_send(fc
, req
);
432 err
= req
->out
.h
.error
;
434 convert_fuse_statfs(buf
, &outarg
.st
);
435 fuse_put_request(fc
, req
);
444 OPT_DEFAULT_PERMISSIONS
,
451 static const match_table_t tokens
= {
453 {OPT_ROOTMODE
, "rootmode=%o"},
454 {OPT_USER_ID
, "user_id=%u"},
455 {OPT_GROUP_ID
, "group_id=%u"},
456 {OPT_DEFAULT_PERMISSIONS
, "default_permissions"},
457 {OPT_ALLOW_OTHER
, "allow_other"},
458 {OPT_MAX_READ
, "max_read=%u"},
459 {OPT_BLKSIZE
, "blksize=%u"},
463 static int parse_fuse_opt(char *opt
, struct fuse_mount_data
*d
, int is_bdev
)
466 memset(d
, 0, sizeof(struct fuse_mount_data
));
468 d
->blksize
= FUSE_DEFAULT_BLKSIZE
;
470 while ((p
= strsep(&opt
, ",")) != NULL
) {
473 substring_t args
[MAX_OPT_ARGS
];
477 token
= match_token(p
, tokens
, args
);
480 if (match_int(&args
[0], &value
))
487 if (match_octal(&args
[0], &value
))
489 if (!fuse_valid_type(value
))
492 d
->rootmode_present
= 1;
496 if (match_int(&args
[0], &value
))
498 d
->user_id
= make_kuid(current_user_ns(), value
);
499 if (!uid_valid(d
->user_id
))
501 d
->user_id_present
= 1;
505 if (match_int(&args
[0], &value
))
507 d
->group_id
= make_kgid(current_user_ns(), value
);
508 if (!gid_valid(d
->group_id
))
510 d
->group_id_present
= 1;
513 case OPT_DEFAULT_PERMISSIONS
:
514 d
->flags
|= FUSE_DEFAULT_PERMISSIONS
;
517 case OPT_ALLOW_OTHER
:
518 d
->flags
|= FUSE_ALLOW_OTHER
;
522 if (match_int(&args
[0], &value
))
528 if (!is_bdev
|| match_int(&args
[0], &value
))
538 if (!d
->fd_present
|| !d
->rootmode_present
||
539 !d
->user_id_present
|| !d
->group_id_present
)
545 static int fuse_show_options(struct seq_file
*m
, struct dentry
*root
)
547 struct super_block
*sb
= root
->d_sb
;
548 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
550 seq_printf(m
, ",user_id=%u", from_kuid_munged(&init_user_ns
, fc
->user_id
));
551 seq_printf(m
, ",group_id=%u", from_kgid_munged(&init_user_ns
, fc
->group_id
));
552 if (fc
->flags
& FUSE_DEFAULT_PERMISSIONS
)
553 seq_puts(m
, ",default_permissions");
554 if (fc
->flags
& FUSE_ALLOW_OTHER
)
555 seq_puts(m
, ",allow_other");
556 if (fc
->max_read
!= ~0)
557 seq_printf(m
, ",max_read=%u", fc
->max_read
);
558 if (sb
->s_bdev
&& sb
->s_blocksize
!= FUSE_DEFAULT_BLKSIZE
)
559 seq_printf(m
, ",blksize=%lu", sb
->s_blocksize
);
563 void fuse_conn_init(struct fuse_conn
*fc
)
565 memset(fc
, 0, sizeof(*fc
));
566 spin_lock_init(&fc
->lock
);
567 mutex_init(&fc
->inst_mutex
);
568 init_rwsem(&fc
->killsb
);
569 atomic_set(&fc
->count
, 1);
570 init_waitqueue_head(&fc
->waitq
);
571 init_waitqueue_head(&fc
->blocked_waitq
);
572 init_waitqueue_head(&fc
->reserved_req_waitq
);
573 INIT_LIST_HEAD(&fc
->pending
);
574 INIT_LIST_HEAD(&fc
->processing
);
575 INIT_LIST_HEAD(&fc
->io
);
576 INIT_LIST_HEAD(&fc
->interrupts
);
577 INIT_LIST_HEAD(&fc
->bg_queue
);
578 INIT_LIST_HEAD(&fc
->entry
);
579 fc
->forget_list_tail
= &fc
->forget_list_head
;
580 atomic_set(&fc
->num_waiting
, 0);
581 fc
->max_background
= FUSE_DEFAULT_MAX_BACKGROUND
;
582 fc
->congestion_threshold
= FUSE_DEFAULT_CONGESTION_THRESHOLD
;
584 fc
->polled_files
= RB_ROOT
;
588 fc
->attr_version
= 1;
589 get_random_bytes(&fc
->scramble_key
, sizeof(fc
->scramble_key
));
591 EXPORT_SYMBOL_GPL(fuse_conn_init
);
593 void fuse_conn_put(struct fuse_conn
*fc
)
595 if (atomic_dec_and_test(&fc
->count
)) {
597 fuse_request_free(fc
->destroy_req
);
598 mutex_destroy(&fc
->inst_mutex
);
602 EXPORT_SYMBOL_GPL(fuse_conn_put
);
604 struct fuse_conn
*fuse_conn_get(struct fuse_conn
*fc
)
606 atomic_inc(&fc
->count
);
609 EXPORT_SYMBOL_GPL(fuse_conn_get
);
611 static struct inode
*fuse_get_root_inode(struct super_block
*sb
, unsigned mode
)
613 struct fuse_attr attr
;
614 memset(&attr
, 0, sizeof(attr
));
617 attr
.ino
= FUSE_ROOT_ID
;
619 return fuse_iget(sb
, 1, 0, &attr
, 0, 0);
622 struct fuse_inode_handle
{
627 static struct dentry
*fuse_get_dentry(struct super_block
*sb
,
628 struct fuse_inode_handle
*handle
)
630 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
632 struct dentry
*entry
;
635 if (handle
->nodeid
== 0)
638 inode
= ilookup5(sb
, handle
->nodeid
, fuse_inode_eq
, &handle
->nodeid
);
640 struct fuse_entry_out outarg
;
643 if (!fc
->export_support
)
648 err
= fuse_lookup_name(sb
, handle
->nodeid
, &name
, &outarg
,
650 if (err
&& err
!= -ENOENT
)
657 if (get_node_id(inode
) != handle
->nodeid
)
661 if (inode
->i_generation
!= handle
->generation
)
664 entry
= d_obtain_alias(inode
);
665 if (!IS_ERR(entry
) && get_node_id(inode
) != FUSE_ROOT_ID
)
666 fuse_invalidate_entry_cache(entry
);
676 static int fuse_encode_fh(struct inode
*inode
, u32
*fh
, int *max_len
,
677 struct inode
*parent
)
679 int len
= parent
? 6 : 3;
683 if (*max_len
< len
) {
685 return FILEID_INVALID
;
688 nodeid
= get_fuse_inode(inode
)->nodeid
;
689 generation
= inode
->i_generation
;
691 fh
[0] = (u32
)(nodeid
>> 32);
692 fh
[1] = (u32
)(nodeid
& 0xffffffff);
696 nodeid
= get_fuse_inode(parent
)->nodeid
;
697 generation
= parent
->i_generation
;
699 fh
[3] = (u32
)(nodeid
>> 32);
700 fh
[4] = (u32
)(nodeid
& 0xffffffff);
705 return parent
? 0x82 : 0x81;
708 static struct dentry
*fuse_fh_to_dentry(struct super_block
*sb
,
709 struct fid
*fid
, int fh_len
, int fh_type
)
711 struct fuse_inode_handle handle
;
713 if ((fh_type
!= 0x81 && fh_type
!= 0x82) || fh_len
< 3)
716 handle
.nodeid
= (u64
) fid
->raw
[0] << 32;
717 handle
.nodeid
|= (u64
) fid
->raw
[1];
718 handle
.generation
= fid
->raw
[2];
719 return fuse_get_dentry(sb
, &handle
);
722 static struct dentry
*fuse_fh_to_parent(struct super_block
*sb
,
723 struct fid
*fid
, int fh_len
, int fh_type
)
725 struct fuse_inode_handle parent
;
727 if (fh_type
!= 0x82 || fh_len
< 6)
730 parent
.nodeid
= (u64
) fid
->raw
[3] << 32;
731 parent
.nodeid
|= (u64
) fid
->raw
[4];
732 parent
.generation
= fid
->raw
[5];
733 return fuse_get_dentry(sb
, &parent
);
736 static struct dentry
*fuse_get_parent(struct dentry
*child
)
738 struct inode
*child_inode
= child
->d_inode
;
739 struct fuse_conn
*fc
= get_fuse_conn(child_inode
);
741 struct dentry
*parent
;
742 struct fuse_entry_out outarg
;
746 if (!fc
->export_support
)
747 return ERR_PTR(-ESTALE
);
751 err
= fuse_lookup_name(child_inode
->i_sb
, get_node_id(child_inode
),
752 &name
, &outarg
, &inode
);
755 return ERR_PTR(-ESTALE
);
759 parent
= d_obtain_alias(inode
);
760 if (!IS_ERR(parent
) && get_node_id(inode
) != FUSE_ROOT_ID
)
761 fuse_invalidate_entry_cache(parent
);
766 static const struct export_operations fuse_export_operations
= {
767 .fh_to_dentry
= fuse_fh_to_dentry
,
768 .fh_to_parent
= fuse_fh_to_parent
,
769 .encode_fh
= fuse_encode_fh
,
770 .get_parent
= fuse_get_parent
,
773 static const struct super_operations fuse_super_operations
= {
774 .alloc_inode
= fuse_alloc_inode
,
775 .destroy_inode
= fuse_destroy_inode
,
776 .evict_inode
= fuse_evict_inode
,
777 .drop_inode
= generic_delete_inode
,
778 .remount_fs
= fuse_remount_fs
,
779 .put_super
= fuse_put_super
,
780 .umount_begin
= fuse_umount_begin
,
781 .statfs
= fuse_statfs
,
782 .show_options
= fuse_show_options
,
785 static void sanitize_global_limit(unsigned *limit
)
788 *limit
= ((totalram_pages
<< PAGE_SHIFT
) >> 13) /
789 sizeof(struct fuse_req
);
791 if (*limit
>= 1 << 16)
792 *limit
= (1 << 16) - 1;
795 static int set_global_limit(const char *val
, struct kernel_param
*kp
)
799 rv
= param_set_uint(val
, kp
);
803 sanitize_global_limit((unsigned *)kp
->arg
);
808 static void process_init_limits(struct fuse_conn
*fc
, struct fuse_init_out
*arg
)
810 int cap_sys_admin
= capable(CAP_SYS_ADMIN
);
815 sanitize_global_limit(&max_user_bgreq
);
816 sanitize_global_limit(&max_user_congthresh
);
818 if (arg
->max_background
) {
819 fc
->max_background
= arg
->max_background
;
821 if (!cap_sys_admin
&& fc
->max_background
> max_user_bgreq
)
822 fc
->max_background
= max_user_bgreq
;
824 if (arg
->congestion_threshold
) {
825 fc
->congestion_threshold
= arg
->congestion_threshold
;
827 if (!cap_sys_admin
&&
828 fc
->congestion_threshold
> max_user_congthresh
)
829 fc
->congestion_threshold
= max_user_congthresh
;
833 static void process_init_reply(struct fuse_conn
*fc
, struct fuse_req
*req
)
835 struct fuse_init_out
*arg
= &req
->misc
.init_out
;
837 if (req
->out
.h
.error
|| arg
->major
!= FUSE_KERNEL_VERSION
)
840 unsigned long ra_pages
;
842 process_init_limits(fc
, arg
);
844 if (arg
->minor
>= 6) {
845 ra_pages
= arg
->max_readahead
/ PAGE_CACHE_SIZE
;
846 if (arg
->flags
& FUSE_ASYNC_READ
)
848 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
850 if (arg
->minor
>= 17) {
851 if (!(arg
->flags
& FUSE_FLOCK_LOCKS
))
854 if (!(arg
->flags
& FUSE_POSIX_LOCKS
))
857 if (arg
->flags
& FUSE_ATOMIC_O_TRUNC
)
858 fc
->atomic_o_trunc
= 1;
859 if (arg
->minor
>= 9) {
860 /* LOOKUP has dependency on proto version */
861 if (arg
->flags
& FUSE_EXPORT_SUPPORT
)
862 fc
->export_support
= 1;
864 if (arg
->flags
& FUSE_BIG_WRITES
)
866 if (arg
->flags
& FUSE_DONT_MASK
)
868 if (arg
->flags
& FUSE_AUTO_INVAL_DATA
)
869 fc
->auto_inval_data
= 1;
870 if (arg
->flags
& FUSE_DO_READDIRPLUS
) {
871 fc
->do_readdirplus
= 1;
872 if (arg
->flags
& FUSE_READDIRPLUS_AUTO
)
873 fc
->readdirplus_auto
= 1;
875 if (arg
->flags
& FUSE_ASYNC_DIO
)
878 ra_pages
= fc
->max_read
/ PAGE_CACHE_SIZE
;
883 fc
->bdi
.ra_pages
= min(fc
->bdi
.ra_pages
, ra_pages
);
884 fc
->minor
= arg
->minor
;
885 fc
->max_write
= arg
->minor
< 5 ? 4096 : arg
->max_write
;
886 fc
->max_write
= max_t(unsigned, 4096, fc
->max_write
);
890 wake_up_all(&fc
->blocked_waitq
);
893 static void fuse_send_init(struct fuse_conn
*fc
, struct fuse_req
*req
)
895 struct fuse_init_in
*arg
= &req
->misc
.init_in
;
897 arg
->major
= FUSE_KERNEL_VERSION
;
898 arg
->minor
= FUSE_KERNEL_MINOR_VERSION
;
899 arg
->max_readahead
= fc
->bdi
.ra_pages
* PAGE_CACHE_SIZE
;
900 arg
->flags
|= FUSE_ASYNC_READ
| FUSE_POSIX_LOCKS
| FUSE_ATOMIC_O_TRUNC
|
901 FUSE_EXPORT_SUPPORT
| FUSE_BIG_WRITES
| FUSE_DONT_MASK
|
902 FUSE_SPLICE_WRITE
| FUSE_SPLICE_MOVE
| FUSE_SPLICE_READ
|
903 FUSE_FLOCK_LOCKS
| FUSE_IOCTL_DIR
| FUSE_AUTO_INVAL_DATA
|
904 FUSE_DO_READDIRPLUS
| FUSE_READDIRPLUS_AUTO
| FUSE_ASYNC_DIO
;
905 req
->in
.h
.opcode
= FUSE_INIT
;
907 req
->in
.args
[0].size
= sizeof(*arg
);
908 req
->in
.args
[0].value
= arg
;
909 req
->out
.numargs
= 1;
910 /* Variable length argument used for backward compatibility
911 with interface version < 7.5. Rest of init_out is zeroed
912 by do_get_request(), so a short reply is not a problem */
914 req
->out
.args
[0].size
= sizeof(struct fuse_init_out
);
915 req
->out
.args
[0].value
= &req
->misc
.init_out
;
916 req
->end
= process_init_reply
;
917 fuse_request_send_background(fc
, req
);
920 static void fuse_free_conn(struct fuse_conn
*fc
)
925 static int fuse_bdi_init(struct fuse_conn
*fc
, struct super_block
*sb
)
929 fc
->bdi
.name
= "fuse";
930 fc
->bdi
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
931 /* fuse does it's own writeback accounting */
932 fc
->bdi
.capabilities
= BDI_CAP_NO_ACCT_WB
;
934 err
= bdi_init(&fc
->bdi
);
938 fc
->bdi_initialized
= 1;
941 err
= bdi_register(&fc
->bdi
, NULL
, "%u:%u-fuseblk",
942 MAJOR(fc
->dev
), MINOR(fc
->dev
));
944 err
= bdi_register_dev(&fc
->bdi
, fc
->dev
);
951 * For a single fuse filesystem use max 1% of dirty +
952 * writeback threshold.
954 * This gives about 1M of write buffer for memory maps on a
955 * machine with 1G and 10% dirty_ratio, which should be more
958 * Privileged users can raise it by writing to
960 * /sys/class/bdi/<bdi>/max_ratio
962 bdi_set_max_ratio(&fc
->bdi
, 1);
967 static int fuse_fill_super(struct super_block
*sb
, void *data
, int silent
)
969 struct fuse_conn
*fc
;
971 struct fuse_mount_data d
;
973 struct dentry
*root_dentry
;
974 struct fuse_req
*init_req
;
976 int is_bdev
= sb
->s_bdev
!= NULL
;
979 if (sb
->s_flags
& MS_MANDLOCK
)
982 sb
->s_flags
&= ~MS_NOSEC
;
984 if (!parse_fuse_opt((char *) data
, &d
, is_bdev
))
990 if (!sb_set_blocksize(sb
, d
.blksize
))
994 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
995 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
997 sb
->s_magic
= FUSE_SUPER_MAGIC
;
998 sb
->s_op
= &fuse_super_operations
;
999 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1000 sb
->s_time_gran
= 1;
1001 sb
->s_export_op
= &fuse_export_operations
;
1008 if ((file
->f_op
!= &fuse_dev_operations
) ||
1009 (file
->f_cred
->user_ns
!= &init_user_ns
))
1012 fc
= kmalloc(sizeof(*fc
), GFP_KERNEL
);
1019 fc
->dev
= sb
->s_dev
;
1021 err
= fuse_bdi_init(fc
, sb
);
1025 sb
->s_bdi
= &fc
->bdi
;
1027 /* Handle umasking inside the fuse code */
1028 if (sb
->s_flags
& MS_POSIXACL
)
1030 sb
->s_flags
|= MS_POSIXACL
;
1032 fc
->release
= fuse_free_conn
;
1033 fc
->flags
= d
.flags
;
1034 fc
->user_id
= d
.user_id
;
1035 fc
->group_id
= d
.group_id
;
1036 fc
->max_read
= max_t(unsigned, 4096, d
.max_read
);
1038 /* Used by get_root_inode() */
1042 root
= fuse_get_root_inode(sb
, d
.rootmode
);
1043 root_dentry
= d_make_root(root
);
1046 /* only now - we want root dentry with NULL ->d_op */
1047 sb
->s_d_op
= &fuse_dentry_operations
;
1049 init_req
= fuse_request_alloc(0);
1052 init_req
->background
= 1;
1055 fc
->destroy_req
= fuse_request_alloc(0);
1056 if (!fc
->destroy_req
)
1057 goto err_free_init_req
;
1060 mutex_lock(&fuse_mutex
);
1062 if (file
->private_data
)
1065 err
= fuse_ctl_add_conn(fc
);
1069 list_add_tail(&fc
->entry
, &fuse_conn_list
);
1070 sb
->s_root
= root_dentry
;
1072 file
->private_data
= fuse_conn_get(fc
);
1073 mutex_unlock(&fuse_mutex
);
1075 * atomic_dec_and_test() in fput() provides the necessary
1076 * memory barrier for file->private_data to be visible on all
1081 fuse_send_init(fc
, init_req
);
1086 mutex_unlock(&fuse_mutex
);
1088 fuse_request_free(init_req
);
1092 fuse_bdi_destroy(fc
);
1100 static struct dentry
*fuse_mount(struct file_system_type
*fs_type
,
1101 int flags
, const char *dev_name
,
1104 return mount_nodev(fs_type
, flags
, raw_data
, fuse_fill_super
);
1107 static void fuse_kill_sb_anon(struct super_block
*sb
)
1109 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1112 down_write(&fc
->killsb
);
1114 up_write(&fc
->killsb
);
1117 kill_anon_super(sb
);
1120 static struct file_system_type fuse_fs_type
= {
1121 .owner
= THIS_MODULE
,
1123 .fs_flags
= FS_HAS_SUBTYPE
,
1124 .mount
= fuse_mount
,
1125 .kill_sb
= fuse_kill_sb_anon
,
1127 MODULE_ALIAS_FS("fuse");
1130 static struct dentry
*fuse_mount_blk(struct file_system_type
*fs_type
,
1131 int flags
, const char *dev_name
,
1134 return mount_bdev(fs_type
, flags
, dev_name
, raw_data
, fuse_fill_super
);
1137 static void fuse_kill_sb_blk(struct super_block
*sb
)
1139 struct fuse_conn
*fc
= get_fuse_conn_super(sb
);
1142 down_write(&fc
->killsb
);
1144 up_write(&fc
->killsb
);
1147 kill_block_super(sb
);
1150 static struct file_system_type fuseblk_fs_type
= {
1151 .owner
= THIS_MODULE
,
1153 .mount
= fuse_mount_blk
,
1154 .kill_sb
= fuse_kill_sb_blk
,
1155 .fs_flags
= FS_REQUIRES_DEV
| FS_HAS_SUBTYPE
,
1157 MODULE_ALIAS_FS("fuseblk");
1159 static inline int register_fuseblk(void)
1161 return register_filesystem(&fuseblk_fs_type
);
1164 static inline void unregister_fuseblk(void)
1166 unregister_filesystem(&fuseblk_fs_type
);
1169 static inline int register_fuseblk(void)
1174 static inline void unregister_fuseblk(void)
1179 static void fuse_inode_init_once(void *foo
)
1181 struct inode
*inode
= foo
;
1183 inode_init_once(inode
);
1186 static int __init
fuse_fs_init(void)
1190 fuse_inode_cachep
= kmem_cache_create("fuse_inode",
1191 sizeof(struct fuse_inode
),
1192 0, SLAB_HWCACHE_ALIGN
,
1193 fuse_inode_init_once
);
1195 if (!fuse_inode_cachep
)
1198 err
= register_fuseblk();
1202 err
= register_filesystem(&fuse_fs_type
);
1209 unregister_fuseblk();
1211 kmem_cache_destroy(fuse_inode_cachep
);
1216 static void fuse_fs_cleanup(void)
1218 unregister_filesystem(&fuse_fs_type
);
1219 unregister_fuseblk();
1222 * Make sure all delayed rcu free inodes are flushed before we
1226 kmem_cache_destroy(fuse_inode_cachep
);
1229 static struct kobject
*fuse_kobj
;
1230 static struct kobject
*connections_kobj
;
1232 static int fuse_sysfs_init(void)
1236 fuse_kobj
= kobject_create_and_add("fuse", fs_kobj
);
1242 connections_kobj
= kobject_create_and_add("connections", fuse_kobj
);
1243 if (!connections_kobj
) {
1245 goto out_fuse_unregister
;
1250 out_fuse_unregister
:
1251 kobject_put(fuse_kobj
);
1256 static void fuse_sysfs_cleanup(void)
1258 kobject_put(connections_kobj
);
1259 kobject_put(fuse_kobj
);
1262 static int __init
fuse_init(void)
1266 printk(KERN_INFO
"fuse init (API version %i.%i)\n",
1267 FUSE_KERNEL_VERSION
, FUSE_KERNEL_MINOR_VERSION
);
1269 INIT_LIST_HEAD(&fuse_conn_list
);
1270 res
= fuse_fs_init();
1274 res
= fuse_dev_init();
1276 goto err_fs_cleanup
;
1278 res
= fuse_sysfs_init();
1280 goto err_dev_cleanup
;
1282 res
= fuse_ctl_init();
1284 goto err_sysfs_cleanup
;
1286 sanitize_global_limit(&max_user_bgreq
);
1287 sanitize_global_limit(&max_user_congthresh
);
1292 fuse_sysfs_cleanup();
1301 static void __exit
fuse_exit(void)
1303 printk(KERN_DEBUG
"fuse exit\n");
1306 fuse_sysfs_cleanup();
1311 module_init(fuse_init
);
1312 module_exit(fuse_exit
);